A model to assess and customize computerized maintenance management systems complies with industry 4.0 vision and requirements: A case study in food processing plant.

(1)

FACULTY OF SCIENCE AND TECHNOLOGY MASTER'S THESIS

Study program:

Offshore Technology - Industrial Asset Management

Spring semester, 2018 (Open access)

Author:

Waqas Ashraf (238225)

……….

(Signature of the author) Program coordinator:

Faculty supervisor: Idriss El-Thalji, External supervisor: Christopher Styles Title:

A model to assess and customize computerized maintenance management systems complies with industry 4.0 vision and requirements: A case study in food processing plant.

Credits: 30 ECTS Keywords:

Computerized maintenance management system, intelligent maintenance, predictive maintenance, Industry 4.0, food processing plant

Number of pages: 102

+ supplemental material/other: 4 Stavanger, 15.06.2018

(2)

ii Abstract

The transformation from Industry 3.0 to Industry 4.0 is rapidly increasing in several industrial sectors where the automated production systems have become more as a cyber-physical and advanced production system. Internet of things (IoT), cyber-physical system, computational intelligence, cognitive capabilities and other disruptive technologies are the key enabling technologies for Industry 4.0. Recent developments in data acquisition techniques i.e. smart sensors has made it possible to initiate changes from the bottom level of complex physical systems.

Cyber-physical systems have become smart enough to interconnect physical and digital world.

Cognitive and computational intelligence capabilities of digital systems have been utilized to control physical equipment with minimum or zero human interference.

The transformation into industry 4.0 vision is quite demanding for several industrial sectors. Transforming the automated asset into a cyber-physical asset is a complex task since the operators need to ensure that their asset is first smart and then develop smart operations in terms of managing the maintenance and executing the required tasks. Smart asset means that the asset e.g. machine should be able to provide/share data about its behavior. Smart maintenance management operations mean that the asset data will be stored, processed, visualized in an automated and cognitive manner by utilizing several disruptive technologies related to automation of work processes e.g. predictive maintenance, artificial intelligence.

From an operator point of view, the computerized enterprise resource planning (ERP) and maintenance management systems (CMMS) e.g. SAP are the most important parts of the entire digital transformation since most of the other systems can be provided by service providers e.g.

algorithms, cloud platforms. However, the operators need to deal with ERP and CMMS on daily bases and almost all acquired data from the asset will be stored and presented via those systems.

CMMS is the core interface between the operator and entire digitalized production system. Thus, the operators shall be enabled to assess which CMMS (as several commercial systems are available) is most cost-effective for their applications, support the completeness of industry 4.0 vision and provide capabilities to be customized.

(3)

iii Concerning these needs, the purpose of this thesis is to determine the key acceptance criteria of selecting CMMS and customization aspects that are required to develop cost-effective CMMS for a specific application.

By keeping in view about requirements, a case study on an electromechanical system has been selected for this thesis where a predictive maintenance program and associated CMMS are developed. Based on this development, the key acceptance criteria and customization aspects for developing cost-effective CMMS are extracted. Moreover, industry 4.0 framework has been applied to upgrade the equipment to meet advanced manufacturing requirements. Research work is carried out on a critical processing equipment named single retention time (SRT) freezing tunnel at Gate Gourmet, United Kingdom. Systems analysis approach has been used to demonstrate case study work.

• I have used systems analysis to analyze how the selected physical system can be retrofitted to be smarter.

• I have developed a Predictive maintenance (PdM) programme for the selected system based on intelligent maintenance layers.

• I have developed a CMMS to support the PdM programme.

• I have extracted the key acceptance criteria of selecting CMMS and customization aspects that are required to develop cost-effective CMMS.

Chapter four provides a clear result of using systems analysis, which is an effective methodology to extract the failure modes, symptoms, and potential process parameters that can be monitored to detect abnormalities. The applied systems analysis of the selected physical system is an effective methodology to extract the failure modes, symptoms (required for health monitoring), and potential process parameters (required for data-driven approach) that can be monitored to detect abnormalities.

The systems analysis of the potential predictive maintenance program highlights the complexity due to the compliance with industry 4.0 requirements of the cyber-physical system at different layers (7 layers). The integrity and work process automation are main requirements to

(4)

iv develop effective predictive maintenance program. Moreover, the developed predictive maintenance program clearly expands the expected functionality of the associated CMMS. It requires CMMS to be smarter (not just a database) by being as user interface integrator between assets (physical machines, human operations) and cyber systems (analytics, transmissions).

The systems analysis of CMMS resulted in identifying the technical functionalities and stakeholders needs with the key acceptance criteria to assess the commercial CMMS and to customize them toward the specific stakeholder's needs. The key acceptance criteria are mainly related to functionality toward increasing data volumes, velocities, veracities, and varieties, visualization, accessibility, privacy, security, interoperability with other systems, usability, scalability, affordability, and completeness to vision.

This thesis can serve as a fundamental guideline of how to analyze your physical systems to identify the requirements for developing smarter assets, predictive maintenance program, and CMMS. The extracted key acceptance criteria for selecting CMMS is significantly important to either select the best commercial option or to customize the existing CMMS solution.

(5)

v

Preface

I am very thankful to my Idriss El-Thalji, associate professor at University of Stavanger, Norway. His passion for maintenance systems has always pushed me to think innovative and out of the box. He has always been a source of inspiration for me. I appreciate his personal support and effort to encourage me to work for my thesis outside the country.

I would also thank my mentor Christopher Styles, head of facilities and fleet UK & Ireland, Gate Gourmet for his support, endless help and generous advice during the research period.

“Finally, I want to acknowledge the support, and love of my family – my parents and siblings. They all kept me going and this work was not possible without their support”.

Author

Waqas Ashraf

(12)

1 1. Introduction

This chapter highlights the thesis background, problem description, scope and its structure.

1.1. Background

Industry 4.0 is a German government initiative, aimed to transform future manufacturing industry in Europe by increasing digitalization and interconnection between products, processes and business models. The transformation from automated and robotic systems (Industry 3.0) to cyber-physical systems is digitally restructuring the whole manufacturing process and value chain (Klitou, Conrads, and Rasmussen, 2017). Cyber-physical systems interconnect physical and digital world. With the cognitive capabilities of the digital world, we can develop smart assets and smart operations to manage advanced manufacturing systems. Smart assets have the functionality of interconnected machines, while smart operations use operational data for artificial intelligence and predictive analytics to improve decision making.

Transformation of automated assets into cyber-physical assets is a complex task. As highlighted in the report of European commission “Germany: Industrie 4.0”, shop floor level involvement is a key barrier in the transformation process (Klitou, Conrads and Rasmussen, 2017).

This digital transformation of smart assets and smart operations will be impossible without the implementation of smart computerized maintenance management system (CMMS), being the critical part of maintenance management system. As cyber-physical systems transform work processes from bottom level and minimize human dependency in the system. Therefore, typical database system and operator-based CMMS systems are not much effective to manage smart systems. It demands to shift maintenance from preventive to predictive approach, where machine failure could be prevented before it happens. This transformation could be carried out by the data- driven approach, in which data collected by smart sensors could be analyzed and utilized for maintenance planning and scheduling. It needs to develop computational and cognitive capabilities in CMMS to use real-time data for maintenance decision making.

(13)

2 1.2. Problem formulation

As per ISO 9000 quality standards, food product quality is very critical control parameter in FMCG industry. Perishable food products quality decrease with increase in temperature. It is very important to maintain, monitor and control food products temperature during the whole supply chain process (harvesting, processing, storage, and delivery). To store food products for a longer period, they need to deep freeze (reduce product temperature below -25 C°) quickly after harvesting.

This deep-freezing process is carried through single retention time (SRT) freezing tunnel which can quickly freeze food products such as fish, shrimp, meat, ice cream, dairy within 2-3- hours cycle. An unscheduled breakdown in processing phase will result in loss of production, loss of quality, labor cost, and spare parts cost. it is very critical to maintain the asset and identify the failure in advance. Another important factor to consider is to carry out repair work inside the freezing tunnel at a temperature below -25 C° is very challenging. It demands maintenance with minimum human involvement. Thus, the purpose of this thesis “is to answer the following research question” (Ekwaro-Osire, Carlos, and Alemayehu, 2018):

How can CMMS system be assessed and customized for the food industry to comply with Industry 4.0 requirements?

In fact, this question leads to three developmental issues: develop smarter assets, smarter predictive maintenance program, and smarter CMMS.

1.3. Limitation/Delimitation

Gate Gourmet is a part of Gate group; one of the world’s largest food processing, airline catering, retail, hospitality, and logistics company. It is operating in four continents and has a worldwide presence with global operations in 60 countries and 160 national & international airports. The company has a total number of 28,000 employees with a net worth of 3.1B CHF. We have selected London Heathrow west production facilities of Gate Gourmet, United Kingdom.

(14)

3 Since this thesis is handling the development as system-level i.e. providing system architecture of PdM program, the detailed technical analyses are not considered. The case study was delimited for only SRT freezing tunnel (within the whole food catering process) in order to manage to demonstrate full case within the limited time (5 months) and resources (available data from vendors and design engineers).

1.4. Structure of the thesis

This thesis has been organized according to guideline provided by the University of Stavanger. The thesis has been divided into five chapters. Chapter 1 highlights the problem, its background, and problem statement. Chapter 2 presents theories related to Industry 4.0, IoT, predictive analytics, and CMMS. Chapter 3 highlights case company, its production facilities, and selected critical system. In chapter 4, I have carried out system analysis of physical systems and CMMS system. In the end, chapter 5 presents discussion and conclusion of the thesis.

(15)

4 2. Literature review and theoretical framework

This chapter highlights theory and concepts related to proposed solution for the problem. It also demonstrates key technologies related to Industry 4.0 vision.

2.1. Industry 4.0

The term Industry 4.0 (fourth industrial revolution) was first demonstrated in Hannover Fair, Germany which emphasis on the transformation of traditional manufacturing systems (Wang and Wang, 2017). In other parts of the world, similar research work is known by different names like smart manufacturing (United States, China) and intelligent manufacturing systems in (Norway, Sweden, and Finland) (Wang and Wang, 2017).

Industry 4.0 is “the extension of traditional manufacturing systems to full integration of physical, emboldened IT system including the internet” (Wang and Wang, 2017). Figure 1 illustrates various stages of industrial evolution from Industry 1.0 to Industry 4.0; transformation from first industrial revolution to fourth industrial revolution (Varghese and Tandur, 2014). It can be seen from the figure that fundamental concepts that lead to the growth in each generation are strongly connected with the critical ideas of automation, self-configuration and self-sufficient systems which have gradually improved over time (Trappey et al., 2016).

Figure 1: Industrial evolution graph. (Trappey et al., 2016)

(16)

5 Industry 4.0 is based on the concept of digitalization of entire value chain; an end to end digital integration of engineering systems and processes (Kagermann, Wahlster, and Helbig, 2013). It develops a networked manufacturing system by using ICT, internet connectivity and communication between devices i.e. machine to machine (M2M) and machines to the computer (M2C) (Wang and Wang, 2017). There are four key components of Industry 4.0.

• Cyber-physical systems (CPS)

• Internet of things (IoT)

• Big data

• Predicative analytics

All of these are interconnected and overlapping with each other. Internet & connectivity are the key enablers for implementation of Industry 4.0. (Wang, 2016)

2.1.2. Cyber-physical systems

Cyber-physical system (CPS) was introduced in the US in 2006. CPS is the integration of physical processes and computational systems. CPS can be considered as a basic building block in the system, which interconnects physical component of a machine (i.e. moving at high speed) with digital systems (smart sensors). Parameters monitoring of a physical process (temperature, pressure, flow, force) is a fundamental requirement in digital transformation. Advancement in the embedded system has helped to increase the functionality of sensors i.e. RFID, smart technology, nanotechnology. Radio-frequency identifications (RFID) sensors have the advantage of non- contact reading and writing capabilities. An RFID sensing system consists of “device (tag), tag reader with an Antenna and transceiver and host system” (for data storage) (Wang and Wang, 2017).

Device tag monitors and stores physical component reading, tag reader communicates the data to host system by non-contact medium (radio wave, microwave). Real-time data acquisition (Information) is the heart of CPS system with integration of communication, computation, and control of physical systems (Wang and Wang, 2017). Figure 2 illustrates key elements of cyber- physical systems.

(17)

6 Figure 2: Cyber-physical systems (Trappey et al., 2016)

RFID technology is suitable for real-time sensing of operating parameters of a machine. It can transmit and receive data for distance up to 12 meters. It has been reported that by 2020, almost 20.8 Billion devices will be using RFID technology for data transmission and communication (Lund et al., 2014).

CPS focuses on the digital part of the manufacturing system. CPS consists of two major functions, (1) real-time data acquisition, interaction, and communication between physical and digital world; (2) translate intelligent computations and cognitive decision (carried out in the digital world) to the physical world. Figure 3 highlights interaction between physical and digital systems.

Figure 3: Physical and digital system interaction (Mueller, Chen and Riedel, 2017)

(18)

7 2.1.3. Internet of things (IoT)

“The term internet of things (IoT) was first coined by the MIT Auto-ID center” (Sethi, Arkko and Keranen, 2012) by Kevin Ashton. It refers to wireless communication between sensors and computing devices through the internet. “IoT can be understood from two perspectives, which are internet-oriented and things-oriented” (Wang and Wang, 2017).; The former one, internet or IP connecting a large number of devices and latter one presents a large number of things (devices).

Each device (things) can be identified with a unique IP address and able to communicate over the internet through a standard communication protocol (Wang and Wang, 2017). With the invention of IPv6, 4.3 Billion devices will be able to connect over the internet with unique IP addresses. The Internet is the major key enabler of IoT concept.

As stated in National Compliance Management Services (NCMS) report that “there is a consensus that linking factory devices to the Internet will become the backbone technology for future manufacturing” (Wu et al., 2015). “IoT is key to improving automation in the manufacturing process” (Wang and Wang, 2018). For example, remote monitoring and control of process equipment and machines. Through cyber-physical systems, IoT allows human and machine to be connected in a whole manufacturing system.

IoT allows distributed cyber-physical systems (RFID sensors) to connect virtually through the internet. It works on the principle of computing concept, in which each physical component can present itself as a digital system and can able to connect, communicate through the internet. It could also identify other devices as well (Auty, 2016). Machine to machine (M2M) and machine to computer (M2C) communications of Smart sensors, wireless sensors, RFID sensors are few examples (Techopedia.com, 2018).

Generic IoT model can be represented by 4C’s, connection, communication, computation, and control. Connection and communication parts are illustrated by cyber-physical systems while computation and control are distinguishing features of IoT. IoT is the technical architecture of cyber-physical systems. In this research work, advanced IoT architecture has been applied which consists of seven layers to implement IoT model in the manufacturing industry. Figure 4

(19)

8 represents the seven layers of IoT and their key functions. It highlights that IoT is a bottom-up approach and systematic deployment architecture for smart manufacturing systems. The function of seven layers of IoT has been explained as under:

Figure 4: IoT layers architecture (Trappey et al., 2016)

Application layer: It consists of physical systems and their components such as; motors, gearbox, conveyors, rotary system.

Perception layer: It consists of measurement sensors and transducers for physical parameters monitoring. These sensors detect a change in different parameters, quantities, and events. These signals are transmitted to PLC system for control and monitoring (Trappey et al., 2016).

Connection/Transmission layer: In this layer, data from sensors is transmitted to data storage service (Trappey et al., 2016). This transmission of data could be wired or wireless. Data is usually stored on cloud servers or on-premises storage facilities.

Conversion layer: Data received from perception layer is not ready for analytics. It needs to be cleaned and processed for better decision making. In conversion layer, data from perception

(20)

9 layers devices is converted into the desired format (structured data) and then it is ready to use for analytics in computation layer. In this step, various types of interfaces are made between

different types of data sources (Trappey et al., 2016). For example, integrating archived and real- time data for specific measurement sensor. Figure 5 illustrates data conversion layer stages. This step carries out pre-processing of data and reduces processing time in computation layer. Data conversion is very critical for real-time applications.

Figure 5: Conversion layer steps. (Trappey et al., 2016)

Computation layer: Computation layer uses machine learning algorithms to analyze the current state of equipment and make future predictions about the condition and behavior of the machine (Trappey et al., 2016).

Cognition layer: “Cognition presents the knowledge gathered in the higher layers for decision support” (Trappey et al., 2016). In cognition layer, computational data is used for monitoring and optimization of process parameters (Trappey et al., 2016).

Configuration layer: “Configuration is the transformation of the intelligence into action (movement from cyberspace to physical space)” (Trappey et al., 2016).

We can utilize the data acquisition, computational and cognitive capabilities of IoT layers for automation of maintenance decision making. For this purpose, an IoT based framework has been presented in Figure 6 for intelligent maintenance system. This framework uses functionalities of the cyber-physical system and IoT architecture for automation of maintenance work process.

These 8 layered structures could be divided into three major sections. The application layer, IIoT layer and cognition, and configuration layer. Application layer interconnects physical world with

(21)

10 the digital world. Industrial internet of things uses computational capabilities of digital world data computation and analytics. While cognition and configuration, layers transform intelligent decision into the physical world.

Figure 6: A Proposed framework for intelligent maintenance system (Trappey et al., 2016; El- Thalji, 2018)

2.1.4. Cloud computing

Cloud technology has been emerged as high performance, low-cost distributed network alternative. It provides hardware virtualization solutions, “virtualization: is a technology that abstracts away the details of physical hardware and provides virtualized resources for high-level applications” (Sakr et al., 2011). It provides high-speed data access and complex computing power for large-scale engineering problems through the internet. The Internet is the backbone of cloud technology (Wang and Wang, 2017).

(22)

11 National Institute of Standards and Technology (NIST) has defined cloud technology as

"cloud computing is a model for enabling ubiquitous, convenient, on-demand network access to a shared pool of configurable computing resources (e.g., networks, servers, storage, applications and services) that can be rapidly provisioned and released with minimal management effort or service provider interaction” (NIST, 2011).

With cloud technology, an enterprise does not need to own expensive hardware and processors (GPU) to solve complex engineering problems. Instead, they can buy cloud access and high-performance computing as service, “hardware as a service (HaaS) and software as a service (SaaS)” (Kumar, 2018). Cloud computing includes hardware storage, cloud computing platform which includes operating systems, program execution environment, database, application development, testing, deployment platform. Famous IT service provider which offer cloud service are Microsoft, Google, and Amazon (Wang and Wang, 2017).

2.1.5. Big data

In manufacturing systems, a huge amount of data is generated by enterprise resources like real-time data from smart and RFID sensors, production system, automation, and control system and ERP/CMMS systems. Through internet connectivity, each component of a physical system could have the ability to connect and communicate on M2M (machine to machine) and M2C (machine to computer) level. This data can be characterized by the 3V’s (volume, velocity, and variety) (Zhang, 2016).

Figure 7 highlights big data sources in maintenance systems. As shown in the figure, the high volume presents data from direct maintenance activities such as; maintenance plans, work orders, condition monitoring data. High velocity indicates high-speed data generated by smart sensors and transducers in real time. High variety represents data from multiple sources with different types and varieties.

(23)

12 Figure 7: Big data sources in maintenance system (Zhang, 2016)

Data from all these sources is huge and heterogeneous, generating from a variety of sources in real time and high speed. This data has characteristics of being complex, decentralized and fast moving. This large amount of data is stored on Cloud servers. With limited human analytical capabilities, it is not possible to analyze the big volume of data.

2.2. Predicative analytics (Big data analytics)

Advancement in high-speed internet, high-performance processors (GPU) and cloud computing capabilities have made possible to store, process and analyze a large amount of data.

Such as data received from condition monitoring sensors which could be used for processing, storage, and analysis for fault diagnosis and prognosis. Big data analytics is also known as predictive analytics. In simple words, predictive analytics is an effective approach to convert insignificant large amount data to meaningful data, which can be used by management for decision making (Wang and Wang, 2017).

Big data analytics is a computational technique for data analysis. It combines historical and real-time data to discover hidden patterns in data sets. It uses machine learning algorithms

(24)

13 to discovers patterns, correlations between various data sources and make a prediction for future events and help in decision making and planning (Wang, 2016).

2.2.1. Machine learning

“Machine learning is a sub-field of artificial intelligence” (Bhandari, 2018) which teaches a machine to learn ‘how to solve a problem’. Arthur Samuel (1959) has described machine learning as a “computer’s ability to learn without being explicitly programmed” (Insights, 2018). Machine learning is different from classical modeling approach. Figure 8 highlights difference between classical & machine learning modeling approach. In machine learning, input (data) and desired outputs (labels) are known and we develop a model by using machine learning algorithm to optimize the outcome.

Figure 8: Classical modeling & machine learning approach. (Tegtmeyer, 2018)

The data is used by the computer to solve a given problem and make future predictions about the event that could be happening in real time (Uz et al., 2018). “Machine learning uses programmed algorithms that receive and analyse input data to predict output values within an acceptable range. As new data is fed to these algorithms, they learn and optimise their operations to improve performance, developing ‘intelligence’ over time” (Wakefield, 2018).

(25)

14 2.2.1.1. Machine learning types

Machine learning can be divided into two major categories; supervised and unsupervised machine learning.

a) Supervised Learning:

Supervised machine learning is analogous to the traditional education system, in which a teacher guides student for the learning process. In supervised learning, inputs (data) and outputs (labels) are known. An algorithm is supposed to find a way to reach to outputs. The algorithm analyzes the data, identify patterns in data, learn from observations.

Supervised machine learning process has been summarized in Figure 9. Algorithms are trained by labeled examples. Training data set guides the algorithm to reach on desired output (Khan, 2018). “This process continues until algorithm achieves a high level of accuracy”

(Wakefield, 2018). The supervisor could be a human, who reviews the accuracy of output. For better accuracy of results, large data sets are required for labeled training data-set.

Figure 9: Supervised machine learning (Van Loon, 2018)

Supervised machine learning can be further classified into three major categories as classification, regression, and forecasting (Nawrocki et al., 2018). In classification, machine learning algorithm classifies input data into categories i.e. Normal and abnormal machine conditions. In regression, machine learning algorithm estimates the relationship between multiple

(26)

15 variables and their dependencies. It is very useful to “predict the remaining useful life of the equipment in real time” (Zhicai, Dongfeng and Xinfa, 2014). In forecasting, future predictions are made on the bases of past and present data. It is very useful for trends analysis.

Figure 10 illustrates supervised machine learning process for predictive maintenance analytics process using supervised learning. Predictive maintenance uses machine learning algorithms to discover hidden patterns and correlation in real time and historical data. A supervised machine learning approach is used to train the model (McDonald, 2018). This model is deployed to predict the failures in the equipment.

Figure 10: Predictive maintenance analytics process. (McDonald, 2018) b) Unsupervised learning

In unsupervised machine learning, “algorithm studies data to identify patterns” (Wakefield, 2018). “There is no answer key or human operator to provide instruction. Instead, the machine determines the correlations and relationships by analysing available data. In an unsupervised learning process, the machine learning algorithm is left to interpret large data sets and address that data accordingly” (Wakefield, 2018).

“The algorithm tries to organise that data in some way to describe its structure. This might mean grouping the data into clusters or arrange it in a way that looks more organised” (Wakefield, 2018). “As it assesses more data, its ability to make decisions on that data gradually improves and

(27)

16 becomes more refined” (Wakefield, 2018). Unsupervised learning is very useful to categorize data and Time series anomalies detection applications. Figure 11 has summarized unsupervised machine learning process.

Figure 11: Unsupervised machine learning process. (Van Loon, 2018) 2.2.1.2. Algorithm selection criteria

There are large numbers of the licensed and open source machine learning algorithms available. Depending on nature of the problem, different types of machine learning algorithm are used. It is important to note that different algorithms have different characteristics, inputs, processing capacity and output accuracy. A single machine learning algorithm cannot solve all types of problems. We may need multiple algorithms to solve different types of problems.

Selection of machine learning algorithm is mainly based on business needs and objective includes prediction requirements, accuracy requirements, time and investment resources (GPU: a graphical processing unit for data computation) available for algorithm training. Size and quality of data is another selection criterion for machine learning algorithm.

Selection of algorithm is often a trial and error method to reach desired output. There are two common methods to verify the algorithms. (1) Using know or recommended algorithms which are already tested to provide the desired output. (2) Using operational data from a process for 3-6 months and fine-tune the algorithm model by using supervised learning. In case of unsupervised machine learning, this duration could be increased like (6-12 months) (Wright, 2018).

2.2.1.3. Artificial neural networks

Artificial neural networks (ANN) are analogous to biological neural networks (neurons).

It defines a “non-linear functional relationship between input and output data” (Graña et al., 2018),

(28)

17 a non-linear function is used to make the relationship between inputs and outputs (Ragab et al., 2016). Figure 12 illustrates general configuration of the artificial neural network. As shown in the figure, “it consists of three layers: the input layer, the hidden layer, and output layer” (Ragab et al., 2016). Input features are linked with an input layer, hidden layer is associated with neuron functions and output (Cerrada et al., 2015). In data-driven models, neural networks are most common algorithms to estimate remaining useful life of a component (Wang and Wang, 2017). It uses supervised learning approach to train a model.

Figure 12: Generic model of artificial neural networks (ANN). (Dymczyk, 2018) 2.3. Predictive maintenance

Predictive maintenance is the industrial application of predictive analytics. Real-time monitoring capability of predictive analytics can help to detect the failure on early stage. It significantly reduces unscheduled maintenance and cost on the equipment. Predictive maintenance helps to predict equipment failure (diagnosis), forecast energy needs, improve operational performance, reduce maintenance cost and improve the reliability of an asset. For example, an irregular pattern in sensor signals can help to predict the failure pattern of an equipment. Any deviation from normal signal behavior can be categorized as a failure (Uk.mathworks.com, 2018).

Predictive maintenance uses cognitive reasoning and makes optimal decisions without human interface. It can provide a future prediction about the failure on very early stage. In

(29)

18 predictive analytics, the health of an asset is determined based on operating conditions and maintenance recommendations based on future failure forecasting. A properly implemented predictive analytics maintenance system “can significantly reduce maintenance cost by reducing unnecessary maintenance” (Jardine, Lin, and Banjevic, 2006) activities and carry out maintenance when it is required. By eliminating unwanted maintenance can reduce asset life cycle cost, unscheduled downtime (Zhang, 2016). Figure 13 highlights the ISO standard procedure for equipment condition monitoring process. It starts with a cost-benefit analysis to implement condition monitoring system. In the first step, critical components, failure modes, and measurement parameters are identified. Followed by fault diagnosis and prognosis.

“Diagnosis is to detect the failure that has occurred in a component (or subsystem) and isolate and identify the root of the failure, based on the data collected by the embedded sensors.

Prognosis is to estimate the time at which a component will fail to operate at its stated specifications based on its current condition as well as the future load and environmental exposure, i.e., the prediction of the remaining useful life (RUL) of the component. RUL is a commonly used parameter to assess the reliability (or reusability) of a used component” (Nee, 2015).

(30)

19 Figure 13: Condition monitoring procedure flowchart (ISO 17359:2011(E))

(31)

20 2.3.1. Predictive maintenance and performance measurement model

Using IoT based intelligent maintenance and predictive analytics capabilities of maintenance system. A predictive maintenance and performance measurement model have been illustrated in Figure 14. As shown in the figure, there are five major modules of intelligent predictive maintenance model. (1) Data acquisition, (2) Data manipulation, (3) Diagnostics and Prognostics, (4) Key performance indicators and (5) Optimization (Wang, 2016; El-Thalji, 2018).

Figure 14: Intelligent predictive maintenance model (Wang, 2016; El-Thalji, 2018) a) Data acquisition

Data acquisition is the first step in the implementation of intelligent predictive maintenance. In this step, appropriate sensors are selected, installed on the machine at a suitable location where optimal output signal can be generated for the condition of the equipment. This data is collected in real time and stored in a cloud server. Then, this data is transformed to a domain which contains maximum information about the condition of the equipment (Wang, 2016).

(32)

21 b) Data manipulation

Data collected from multiple sensors are not readily available for analytics, it has some missing features such as noise, redundant data, incorrect sensors readings. Therefore, it is necessary to sort, filter and prioritize the raw data before processing (Lee et al., 2014). In data manipulation module, two steps are carried out. (1) Pre-processing and conditioning and (2) feature extraction. In pre-processing and signal conditioning: signal characteristics and quality are improved. Various techniques are used for this purpose like filtration, amplification, data compression, de-noising; to remove noise from the signal to improve the signal-to-noise ratio. In features extraction: important features from the pre-processed signal are extracted which highlight incipient failure (Wang, 2016). Generally, there are three domains, from which we can “extract features time domain, frequency domain, and time-frequency domain” (Ahmed, Banaee and Loutfi, 2013). Selection of domain depends on equipment system analysis (Wang, 2016).

c) Diagnostics and prognostics

Signal identified in data manipulation section is used to identify faults of equipment.

“Diagnostics focuses on detection, isolation, and identification of fault” (Wang, 2016) while prognostics focuses on the prediction of occurring of a fault (in future). Various types of fault diagnosis and equipment condition assessment models are available which could be selected based on system analysis. Selection of model is based on the availability of historical data and learning techniques; supervised and unsupervised (Uk.mathworks.com, 2018). Figure 15 highlights method for fault diagnostics and prognostics.

(33)

22 Figure 15: Fault diagnostics and prognostics model (Wang, 2016)

Time is a critical factor in prognosis. Remaining time to failure of a component (“how much time is left before a failure occurs” (Wang, 2016) is calculated. It is also called remaining useful life (RUL) of a component or machine. Intelligent predictive maintenance system calculates the remaining useful life of a component based on the data-driven model. It also analyzes the relationship between remaining useful life (RUL) and condition of the component or machine (Wang, 2016).

d) Key performance indicators

Degradation of components is graphically presented by using a radar chart/spider chart or risk chart. Key performance indicators are scaled on the basis of severity, criticality, business rules and safety. These charts help operators to visualize condition of the components graphically.

e) Optimization

It is possible to design a highly reliable equipment, but it may not be economical for industrial use. As high reliability comes up with a high cost. It is important to choose an optimal

(34)

23 solution between failure, cost, and reliability of equipment. Similarly, in intelligent predictive maintenance, it is possible to set maintenance diagnosis and prognosis rules and limits with respect to failure, cost or reliability. Each factor has its own consequences, high-reliability results in low equipment failure but high maintenance cost.

The best approach is to make a trade-off between equipment failure, reliability, and maintenance cost. Maintenance rescheduling should be carried out at an optimal interval where lowest cost results in low equipment failure and high performance and reliability. A conceptual idea has been presented in Figure 16 to highlight the relationship between failure, reliability and maintenance cost.

Figure 16: Relationship between failure time, reliability, and maintenance cost (Wang et al., 2015)

2.4. Computerized maintenance management system (CMMS)

“A computer-managed maintenance system is an integrated set of computer programs and data files designed to provide its user with a cost-effective means of managing massive amounts of maintenance, inventory control, and purchasing data” (Cato and Mobley, 2002). It is important to note that “the CMMS is a tool used to improve maintenance and related activities”, “it does not

(35)

24 manage the maintenance operations” (Cato and Mobley, 2002). It is also referred to as enterprise asset management software (EAM).

Computerized maintenance management system (CMMS) is a major part of the maintenance information management system (en.wikipedia.org, 2018). It’s a software platform for maintenance system management. It helps in systematic planning, execution, and control of maintenance activities. It provides a cost-effective way to manage human and capital resources (Cato and Mobley, 2002). Figure 17 highlights the resources which need to be managed in an enterprise. It can be seen from the figure that maintenance is the sub-part of enterprise and production system. Maintenance systems have high importance as both depend on it. CMMS software helps to manage resources such as labor, spares, tools, information, cost and out-sourced repair activities. These inputs result in production output, availability, maintainability and the safety of assets.

Figure 17: Input and output model for an enterprise. (Al-Turki, 2009)

CMMS is a transformation from a paper-based working environment to a computerized digital storage. It helps to eliminate traceability and recording of paperwork. Table 1 summarizes the key benefits of CMMS. CMMS helps to manage track maintenance activities such as percentage of PM work completed and backlog work order management.

(36)

25 Table 1: Benefits of CMMS (Ahmed Soliman, 2015)

Computerized maintenance management system (CMMS)

• Reduce maintenance backlog

• Reduce maintenance cost

• Reduce overtime

• Reduce follow up a time to repairs

• Reduce outsourced contract maintenance work

• Improve maintenance planning and scheduling

• Improve maintenance service tracking

• Improve technician and service engineer performance

• Improve technician and supervisor planning

• Very helpful in various certifications (ISO/food/BRC)

Based on the functionality of CMMS, the software package is “grouped into subsystem or modules for specific acidity set. These subsystems may include but are not limited to” (Cato and Mobley, 2002).

• Equipment/asset register

• Preventive maintenance (PM) planning

• Work order management

• Human resource management

• Inventory management

Equipment/asset register is a database of all the equipment in the plant. All maintenance activities are linked to an asset. So, it is linked to all other databases such as PM planning, inventory, and purchase. Preventive maintenance (PM) planning module contains PM plan for the registered assets/ equipment based on maintenance checklist and frequency. The work order is the backbone of CMMS system. (Cato and Mobley, 2002). A work order can be generated by planned PM or unplanned maintenance activity. As shown in Figure 18; work order is the heart of CMMS system. All modules are connected and updated through a work order. Human resources include personnel and technician who carry out maintenance work activity. Inventory management includes warehousing, purchase, and ordering of spare parts (Cato and Mobley, 2002).

(37)

26 Figure 18: Integration of various CMMS modules. (Ahmed Soliman, 2015)

CMMS software is available in a variety of cost-effective models such as LAN-based or licensed software or web-based or software as-as service (SaaS) model. The LAN-based software is purchased by the company and all the data is stored on the company servers, major maintenance of parts is carried out by company IT staff or with the help of service vendor. Web-based software is available through monthly or yearly subscription. Data is stored in cloud storage and all maintenance activates are carried out by the software provider. In next decade, LAN-based CMMS software will be preferred by large organizations while the small organization will prefer SaaS- based solution. (Steenstrup and Analyst, 2017).

With the significant improvements in ICT system and high-speed internet. Focus on tradition CMMS just as the database is reducing. Instead, it's shifting toward automation of decision making using analytic tools. Based on current and future market trends, innovation and addressing customer needs (Steenstrup and Analyst, 2017), New trends in CMMS or enterprise asset management industry are (Khan, 2018)

• Industrial internet of things (IIoT)

• Cloud computing

• Big data and advanced analytics

(38)

27

• Blockchain

• Mobility and SaaS

Industrial internet of things (IIoT) uses industrial Ethernet for connected devices concept.

Cloud computing and big data analytics include processing of a large amount of data for predictive analytics using machine learning algorithms. Blockchain technology is helping asset management to capture all transactions and change in the state of the equipment (Venkataraman et al., 2017).

CMMS software vendors can be categorized into two major groups tier 1 and tier 2. Tier 1 CMMS software includes SAP, IBM Maximo, Infor LN and IFS. While tier 2 CMMS can be named such as Avantis (Schneider Electric), Ellipse (ABB), Oracle, Mainsaver, eMaint, Fiix etc.

There are many CMMS software vendors who are working to introduce new technicities in asset management industry. Figure 19 highlights the Gartner 2017 report, which has scaled famous CMMS software such as Infor, IBM, SAP, and IFS based on their vision and ability to execute the changing market condition (Steenstrup and Analyst, 2017).

Figure 19: Magic quadrant of enterprise asset management software. (Steenstrup and Analyst, 2017)

(39)

28 2.4.1. Needs, requirements and acceptance criteria for CMMS system

To identify CMMS system needs and requirements, I have carried out the interview with maintenance manager, production manager, maintenance supervisors and technicians in the company. I have used a template sheet to record customer response. This sheet is attached in Annex A. Feedback received from the management have been summarized in Table 2.

Table 2: Customers’ needs and requirements

Stakeholder Needs Requirements

Maintenance The company needs a maintenance management system to solve their business needs

The system should be able to capture maintenance data for all assets.

Maintenance

The system should record all the maintenance information of asset, stakeholders, and shareholders and the strong relations with company other software.

The system should work with a central focus to obtain all information to maintenance managers.

Maintenance

The system should be able to recognize equipment problems and report them in a timely manner.

The system should be able to detect errors, failure, and correct equipment problems.

Maintenance

Software should have the latest information about the condition of the equipment.

Software should have access from real-time data from machine sensors.

Equipment and software should be able to send and receive

information.

Machine condition data should be stored in such a way that software can access it.

Maintenance The information should be sent to software automatically.

The software should make data processing, (data received from sensors).

The system should be able to transfer important data from the equipment quickly.

Maintenance The system should be able to detect all sensors.

The software should be able to collect data from multiple data sources, types and formats.

The system should be able to store and process a large amount of data.

Maintenance The system should work without human interaction.

The software should work in automatic mode.

Maintenance All technician should be able to use the system.

There should be enough number of the user account in the software.

(40)

29 All technician should be trained to use the software.

Maintenance The output from the software should be understandable.

The output data should be presented in an in comprehensive and summarized way.

Maintenance The system should ensure data privacy.

The system should be secure to operate and use.

From stakeholder needs and requirements, we can deduce following acceptance criteria’s.

The CMMS software should be capable to handle maintenance data from various sources e.g.

online, offline, real-time sensor data, equipment current condition data, equipment failure historical data. It should be able to receive real-time data from the equipment sensors. The software should be able to store big data. Which could be achieved using cloud storage of on- premise data storage. For automatic operations, we need to implement predictive analytics approach for automatic decision making without human interface. It should be able to support intelligent predictive maintenance and performance measurement (PdM). The output of CMMS should be presented and visualized in a summarized way, reports, charts and real-time alarms.

(41)

30 3. Case study and data collection

This chapter elaborates the summary of fieldwork carried out at case company. Its highlights its production, operational facilities, and analysis of the critical physical system.

3.1. Case company

Gate Gourmet is one of the world largest food processing, airline catering and food provisioning services company established in 1992 and currently operating is four continents viz.

Europe, America, Australia, Asia. Gate Gourmet has a worldwide presence with global operations in 60 countries and 160 national & international airports. The company is specialized in food processing and catering service for airplane flights with a total number of 28,000 employees with a net worth of 3.1B CHF. The companies main headquarter are located in Zurich, Switzerland. The company provides 250 Million meals per year to more than 260 airline companies at 120 local and international airports (Airline Suppliers, 2018).

In the airline industry, food supply-demand is continuously changing due to new ticket booking, cancelation, and delay in flight schedule. The company uses SCALA software to manage its food supply demands. More than 250 airlines companies scan interact with Gate Gourmet through SCALA system. They can easily request desired on of meals, modify and update their food demand even some hours before the flight (Coursehero.com, 2018). Figure 20 represents Gate Gourmet supply chain process. It consists of a complete set of activities from the purchase of raw materials, processing, freezing, and delivery of end product to the consumers.

Figure 20: Processing plant - Supply chain process

(42)

31 Raw material receiving involves the purchase of raw material of aquaculture products (shrimps, fish) and meat. This material is cleaned and washed and sorted according to size and quality of materials. Later, these materials are treated with SMBS (Sodium metabisulfite) to increase the shelf life of the product. The product is quickly freezing to achieve a core temperature of -18 C to protect from bacterial growth. Finally, this is packaged and stored in cold storage.

3.2. Production facilities

Gate Gourmet production facility is located near to London Heathrow airport and it provides services to airline companies at Heathrow airport, the 7^th busiest airport in the world (en.wikipedia.org, 2018). It has large food processing plant with several production and process sections. Raw material receiving area, processing areas, freezing system, packing stations, cold storage and shipping areas. Freezing systems consists of three large-scale freezers, SRT freezing tunnel 1, SRT freezing tunnel 2 and an individual quick (IQF) freezer; which freeze the product individually. SRT freezing tunnel freezes products in a minimum of one kg box, which is called slab. Such low temperature is achieved by using Ammonia gas for the refrigeration cycle. A separate refrigeration plant is installed close to the processing plant to supply chilled air.

3.2.1. Operating facilities

Gate Gourmet day to day operations is managed by four main departments, production, quality, process control and maintenance dept. Production dept. deals with production requirements and production targets as per sales demand. Quality dept. look after all the products are produced as per SOP’s quality requirements. Hygiene and sanitation are responsible to maintain hygienic conditions in the processing facility. Process control dept. keeps an eye on process parameters for temperatures, chemical ratios etc. Maintenance department takes care of all equipment operations, maintenance, and repair. Figure 21 highlights Gate Gourmet production facilities.

(43)

32 Figure 21: Gate Gourmet – operational facility

3.3. Selected critical physical system

SRT freezing tunnel is designed, developed and manufactured by Hense Jensen Engineering, Denmark. Freezing tunnel/Rack freezer is a large-scale industrial freezer which is used to quickly freeze bulk quantity food products (shrimps, fish, meat, dairy ice cream etc.). The freezing tunnel is a critical equipment is food processing industry. It works on the principle of single retention time (SRT). Retention time is the amount of time a product spends in the freezer once it has been injected into the freezer. This time varies with respect to product size, thinness, and shape. For example, the retention time for large products (fish) is different from small ones (shrimp).

Figure 22 illustrates overall system view SRT freezing tunnel. As shown in the figure, it consists of infeed system, rack system, and outfeed system. Infeed system consists of infeed conveyor, infeed pusher, and buffer plate. Rack system consists of 216 racks, chain tension station.

The outfeed system consists of outfeed buffer plate, pusher, and outfeed conveyor. It also consists of electromechanical and refrigeration system.

(44)

33 Figure 22: Single retention time (SRT) Freezing tunnel (Hans Jensen Engineering, 2018)

The SRT freezing tunnel “freezes the same product at a time as the tunnel operates with

‘First IN – First OUT’, which means the products get the same retention time. This tunnel is for small to large quantities of products per hour” (Hans Jensen Engineering, 2018). SRT freezing tunnel has a production capacity of 5.1 tons in one complete cycle. A complete cycle lasts for 2-3 hours, depending on the size of the product. Large products need higher freezing time to achieve core temperature. Table 3 presets production capacity calculation for SRT freezing tunnel.

Table 3: SRT freezing tunnel production capacity calculations SRT freezing tunnel production capacity

1 slab weight 1 kg

No of slabs on a rack 24 (8 slabs in a row x 3 row on one

rack) Total Product weight on one rack 24 kg

Total no. of racks 216

Total product processed in one cycle 5184 kg (24 x 216)

Processing capacity (1 Batch) 5.1 Ton/cycle* *one cycle is approx. 2-3 hours Freezing tunnel Processing capacity 40 Ton/Day

(45)

34 4. System analysis and results

This chapter demonstrates applied system analysis approach to prose solution for the selected physical system and CMMS systems.

4.1. System analysis for selected physical system 4.1.1. Lifecycle processes

SRT freezing tunnel has eight life cycle processes which are listed in Table 4. Stakeholders on lifecycle phases are distributed between vendor and customer. Hans Jensen Engineering, Denmark is the vendor which have designed and manufacture the system. Gate Gourmet was the customer to purchase the equipment and responsible for the operation and maintenance of the equipment.

Table 4: Freezing tunnel lifecycle process

Sr. # Lifecycle process

Stakeholder Vendors:

Hans Jensen Engineering

Customer:

Gate Gourmet

1 Research & development ✓ -

2 Manufacture ✓ -

3 Logistics & supply ✓ ✓

4 Installation & Commissioning ✓ ✓

5 Training ✓ ✓

6 Operation & Maintenance - ✓

7 Upgradation & modification ✓ ✓

8 Disposal & retirement ✓ ✓

4.1.1.1. Key stakeholders

There are three major stakeholders of SRT freezing tunnel, production, maintenance and quality department. The production department is responsible for utilization of freezing tunnel.

The maintenance department is responsible for daily operations and maintenance of freezing

(46)

35 tunnel while the quality department is responsible to ensure that product processed in SRT freezing tunnel is as per company quality requirements.

4.1.2. System context

SRT freezing tunnel consists of three major parts, infeed, rack system and outfeed system.

It also consists of mechanical, refrigeration and electrical system. Figure 23 illustrates system context of SRT freezing tunnel.

Figure 23: System context of Freezing tunnel (Hans Jensen Engineering, 2018) 4.1.3. System breakdown

SRT Freezing tunnel consists of three major systems, (1) Infeed system, (2) Rack system and (3) Outfeed system. Infeed system consists of 2 infeed conveyors, photo sensor, buffer plate, slabs pusher (wiper) Safety glass cover. Photosensor counts 8 product slabs and stops the conveyor. Product slabs are moved from the conveyor to racks by means of the automatic pusher

(47)

36 (sweeper). Slabs pusher moves the 8 slabs to buffer plate (from infeed conveyor). When buffer plate and rack are on the same level, slabs pusher moves 8 product slabs to racks. Figure 24 presents SRT freezing tunnel infeed system.

Figure 24: Infeed system - SRT freezing tunnel. (Hans Jensen Engineering, 2018)

Rack system consists of racks, main chain, and two drive motors, chain tension stations (lower, middle and top), bearings, wheels and refrigeration system (chilled air circulation fans, evaporators and defrost system). Racks are driven by two main chains, which are installed on each side of racks. Size of each chain is 3 inch (thickness). Racks are connected with a chain through chain pins. Broken chain pins can cause a breakdown. Racks are turned through PEHD wheels and always kept in the horizontal direction. Figure 25 highlights inside view of SRT freezing tunnel.

(48)

37 Figure 25: Inside view - SRT freezing tunnel. (Hans Jensen Engineering, 2018)

There are 216 racks in the freezing tunnel. Each rack holds three rows of product cartons and in each row contains eight slabs/carton of product. These slabs move on racks for almost 2 – 3 hours, unless core temperature of the product is achieved (-18 ^oC). Figure 26 highlights rack system of SRT freezing tunnel.

Figure 26: Racks system - Freezing tunnel. (Hans Jensen Engineering, 2018)

A model to assess and customize computerized maintenance management systems complies with industry 4.0 vision and requirements: A case study in food processing plant.

FACULTY OF SCIENCE AND TECHNOLOGY MASTER'S THESIS

Table of Contents

Preface